Anomaly Detection for Time Series Data from Internet of Things Devices
Raza, Syed Shamir (2024)
2024
Tietojenkäsittelyopin maisteriohjelma - Master's Programme in Computer Science
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
Hyväksymispäivämäärä
2024-12-08
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2024120210689
https://urn.fi/URN:NBN:fi:tuni-2024120210689
Tiivistelmä
The rapid growth of Internet of Things (IoT) devices resulted in an unprecedented generation of vast amounts of time series data, in terms of volume, variety, and velocity, known as big data. This data is mainly for monitoring, control, process optimization, and data-driven decision-making in industries. Therefore, identifying anomalies within the data is of vital importance. The anomalies can signify operational inefficiencies, equipment malfunctions, and cyber threats and pose significant challenges. This thesis provides a comprehensive study and approach for anomaly detection of time series data generated from IoT sensors of Computer Numerical
Control (CNC) machines.
This thesis explores and evaluates various preprocessing techniques, feature extraction, and feature transformation to make the raw data ready to be trained by machine learning (ML) algorithms, focusing on accurately detecting anomalies for a manufacturing facility. Anomaly detection works by detecting or picking up patterns in data points that behave abnormally with the normal data points or from vast data observations. Faulty procedures, faulty IoT sensors, or network latency could be possible reasons for anomalies in data. The thesis addresses the challenges of building intelligent detection systems and presents a viable way to detect anomalies from the IoT stream.
The approach starts with challenges inherently present in the IoT time series data, which includes noise and missing values. This thesis proposes a robust strategy for designing a data preprocessing pipeline to improve the data quality, which can aid in accurately detecting anomalies by ML techniques. Different ML-based anomaly detection techniques are used, which include Multi-Layer Perceptron (MLP), Recurrent Neural Network - Long Short Term Memory (RNN-LSTM), Convolutional Neural Networks (CNN), and Extreme Gradient Boosting (XGBoost).
Anomaly detection is a big challenge, with numerous real-world use cases and vast practical applications. By leveraging techniques discussed in the thesis, organizations can detect real-time anomalies for timely failure prevention. This thesis contributes to the ongoing research in the growing field of detecting anomalies in the big data generated from IoT devices with applications across different industries.
Control (CNC) machines.
This thesis explores and evaluates various preprocessing techniques, feature extraction, and feature transformation to make the raw data ready to be trained by machine learning (ML) algorithms, focusing on accurately detecting anomalies for a manufacturing facility. Anomaly detection works by detecting or picking up patterns in data points that behave abnormally with the normal data points or from vast data observations. Faulty procedures, faulty IoT sensors, or network latency could be possible reasons for anomalies in data. The thesis addresses the challenges of building intelligent detection systems and presents a viable way to detect anomalies from the IoT stream.
The approach starts with challenges inherently present in the IoT time series data, which includes noise and missing values. This thesis proposes a robust strategy for designing a data preprocessing pipeline to improve the data quality, which can aid in accurately detecting anomalies by ML techniques. Different ML-based anomaly detection techniques are used, which include Multi-Layer Perceptron (MLP), Recurrent Neural Network - Long Short Term Memory (RNN-LSTM), Convolutional Neural Networks (CNN), and Extreme Gradient Boosting (XGBoost).
Anomaly detection is a big challenge, with numerous real-world use cases and vast practical applications. By leveraging techniques discussed in the thesis, organizations can detect real-time anomalies for timely failure prevention. This thesis contributes to the ongoing research in the growing field of detecting anomalies in the big data generated from IoT devices with applications across different industries.